Updated on 2024-11-11 GMT+08:00

Volcano Scheduler Release History

Table 1 Release history

Add-on Version

Supported Cluster Version

New Feature

1.13.3

v1.21

v1.23

v1.25

v1.27

v1.28

v1.29

  • Supported scale-in of customized resources based on node priorities.
  • Optimized the association between preemption and node scale-out.

1.13.1

v1.21

v1.23

v1.25

v1.27

v1.28

v1.29

Optimized scheduler memory usage.

1.12.18

v1.21

v1.23

v1.25

v1.27

v1.28

v1.29

  • CCE clusters 1.29 are supported.
  • The preemption function is enabled by default.

1.11.21

v1.19.16

v1.21

v1.23

v1.25

v1.27

v1.28

  • Supported Kubernetes 1.28.
  • Supported load-aware scheduling.
  • Updated image OS to HCE 2.0.
  • Optimized CSI resource preemption.
  • Optimized load-aware rescheduling.
  • Optimized preemption in hybrid deployment scenarios.

1.11.6

v1.19.16

v1.21

v1.23

v1.25

v1.27

  • Supported Kubernetes 1.27.
  • Supported rescheduling.
  • Supported affinity scheduling of nodes in the node pool.
  • Optimized the scheduling performance.

1.10.7

v1.19.16

v1.21

v1.23

v1.25

Fixes the issue that the local PV add-on fails to calculate the number of pods pre-bound to the node.

1.10.5

v1.19.16

v1.21

v1.23

v1.25

  • The volcano agent supports resource oversubscription.
  • Adds the verification admission for GPUs. The value of nvidia.com/gpu must be less than 1 or a positive integer, and the value of volcano.sh/gpu-core.percentage must be less than 100 and a multiple of 5.
  • Fixes the issue that pod scheduling is slow after PVC binding fails.
  • Fixes the issue that newly added pods cannot run when there are terminating pods on a node for a long time.
  • Fixes the issue that volcano restarts when creating or mounting PVCs to pods.

1.9.1

v1.19.16

v1.21

v1.23

v1.25

  • Fixes the issue that the counting pipeline pod of the networkresource add-on occupies supplementary network interfaces (Sub-ENI).
  • Fixes the issue where the binpack add-on scores nodes with insufficient resources.
  • Fixes the issue of processing resources in the pod with unknown end status.
  • Optimizes event output.
  • Supports HA deployment by default.

1.7.2

v1.19.16

v1.21

v1.23

v1.25

  • Adapts to clusters 1.25.
  • Improves scheduling performance of volcano.

1.7.1

v1.19.16

v1.21

v1.23

v1.25

Adapts to clusters 1.25.

1.4.7

v1.15

v1.17

v1.19

v1.21

Deletes the pod status Undetermined to adapt to cluster Autoscaler.

1.4.5

v1.17

v1.19

v1.21

Changes the deployment mode of volcano-scheduler from statefulset to deployment, and fixes the issue that pods cannot be automatically migrated when the node is abnormal.

1.4.2

v1.15

v1.17

v1.19

v1.21

  • Resolves the issue that cross-GPU allocation fails.
  • Supports the updated EAS API.

1.3.7

v1.15

v1.17

v1.19

v1.21

  • Supports hybrid deployment of online and offline jobs and resource oversubscription.
  • Optimizes the scheduling throughput for clusters.
  • Fixes the issue where the scheduler panics in certain scenarios.
  • Fixes the issue that the volumes.secret verification of the volcano job in the CCE clusters 1.15 fails.
  • Fixes the issue that jobs fail to be scheduled when volumes are mounted.

1.3.3

v1.15

v1.17

v1.19

v1.21

Fixes the scheduler crash caused by GPU exceptions and the privileged init container admission failure.

1.3.1

v1.15

v1.17

v1.19

  • Upgrades the volcano framework to the latest version.
  • Supported Kubernetes 1.19.
  • Adds the numa-aware add-on.
  • Fixes the deployment scaling issue in the multi-queue scenario.
  • Adjusts the algorithm add-on enabled by default.

1.2.5

v1.15

v1.17

v1.19

  • Fixes the OutOfcpu issue in some scenarios.
  • Fixes the issue that pods cannot be scheduled when some capabilities are set for a queue.
  • Makes the log time of the volcano component consistent with the system time.
  • Fixes the issue of preemption between multiple queues.
  • Fixes the issue that the result of the ioaware add-on does not meet the expectation in some extreme scenarios.
  • Supports hybrid clusters.

1.2.3

v1.15

v1.17

v1.19

  • Fixes the training task OOM issue caused by insufficient precision.
  • Fixes the GPU scheduling issue in CCE v1.15 and later versions. Rolling upgrade of CCE versions during task distribution is not supported.
  • Fixes the issue where the queue status is unknown in certain scenarios.
  • Fixes the issue where a panic occurs when a PVC is mounted to a job in a specific scenario.
  • Fixes the issue that decimals cannot be configured for GPU jobs.
  • Adds the ioaware add-on.
  • Adds the ring controller.